.

[. @_OMaas]

@8teAPi Have you used Claude code? It’s broken. None of this slop works

[. @_OMaas]

@claudeai Make auto mode work first guys?

[. @_OMaas]

@bcherny @demirelo The auto classifier hasn't worked for weeks, with no ability to escalate and permit bash usage. The model will then try to find other ways to use bash. This seems not a great thing.

[. @_OMaas]

@karpathy https://t.co/cLXFFNUwPz 🤮🤮🤮🤮🤮🤮🤮

[elie @eliebakouch]

mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community

also the fact that this is un purpose not visible to the user is crazy

[. @_OMaas]

@bcherny auto-stupid on LLM-shaped stuff is v poor without observability. It's nefarious and deeply disappointing.

[. @_OMaas]

@noctus91 @_akhaliq @NVIDIAAI Obviously worse than moondream

[. @_OMaas]

@kalomaze transforms now are mostly conditional operators whose behavior depends on a learned sequence representation, transformer or not. the field is mostly dead anyway and for video it didn’t work ¯\_(ツ)_/¯

[. @_OMaas]

@norpadon Sadly not a lot better :/

[. @_OMaas]

@MsiDouglas Bro have you seen most of the people they’ve hired

[. @_OMaas]

@RichardHanania This is back when h1bs were materially different, pre co-option by mediocre talent

[. @_OMaas]

@AlexNowrasteh H1b being highly skilled lolololol

[. @_OMaas]

@JosephPolitano The current h1b system is badly broken and needs reform. It could be greatly optimized

[. @_OMaas]

@Hesamation The code is shit

[. @_OMaas]

@MsiDouglas The other data people we hired are terrible

[. @_OMaas]

@CharIieBennett Tbf teams who want to win everything aren’t buying Elanga

[. @_OMaas]

@Edit_Kev Q is how do we spend it? Clear to me talent ID has become … less good

[. @_OMaas]

@MsiDouglas Wissa has 1+1 on his contract.

[. @_OMaas]

@RorySmith It’s not a dichotomy. There’s a better solution than this

[. @_OMaas]

@GeordieMAG_ Obviously not

[. @_OMaas]

@MsiDouglas Clearly fire whoever was in charge of whatever this has been